Telocracy: Joe Edelman on AI that holds institutions to their purposes
Telocracy: Joe Edelman on AI that holds institutions to their purposes
An essay proposing that institutions be bound to written, revisable statements of purpose, with AI interpreting and enforcing them, and that people be able to form small purpose-based groups easily. Published as No. 7 of Pax Machina, a new venue for proposals about institutions in a world with powerful AI.
Rufus's note (2026-11-05): Joe Edelman essay. Needs some thorough checking. Also a new journal. — Rufus Pollock
Links
- Essay: Telocracy, Joe Edelman, 3 Nov 2026 (about 30 minutes; intro by Nora Ammann)
- Appendices: https://paxmachina.ai/telocracy/appendices
- Pax Machina: https://paxmachina.ai/ (about)
The argument in brief
Edelman's opening line: "Telocracy is a vision for a society where AI holds institutions accountable to their purposes". The test for an institution becomes "whether an institution remains faithful to its articulated purpose", not whether it pleases shareholders or voters (stakeholder governance) or follows its rules (procedural governance).
The main proposals:
- Purpose documents. Each institution has a constitution-like statement of purpose, by analogy with Claude's constitution. AI models interpret it, audit fidelity and adjudicate disputes. Precedents: the Rule of St Benedict, US public-benefit corporations, charities.
- Deepening, not drifting. Purposes can be revised, but only by "deepening", not swapping one aim for another. Successors "inherit something they can revise, but not appropriate."
- Purpose-fit groups. AI matches people into lightweight associations around shared purposes, giving a "values choice" alongside consumer choice.
- Purpose-bound property. Brands, land or compute licensed on condition they serve a purpose that is "plausibly positive-sum".
- Joint-outcome markets. Groups pool money to pay for a shared result, for example widowed members each seeing three people a week.
- Government. Agencies decide cases against stated mandates, with reasoning anyone can check.
He notes it "doesn't require powerful AI, but it helps a lot."
First critique (needs the thorough check)
- Who checks the AI? AI interprets and enforces the purposes, but the essay's own examples show the model choosing between readings. How that judgment is audited is left open, and the metric-gaming problem it aims to fix can recur inside the model.
- Who decides "deepening"? The line between deepening and substitution is a judgment call made by review panels. Capture is acknowledged, not solved.
- "Plausibly positive-sum" is vague. The essay admits many legitimate trades leave someone worse off.
- The evidence is thin. Much of the support is near-future fiction (a dating app, a widowers' pool, kitchens). A few academic studies are cited for AI-assisted matching; the "Cambrian explosion" of associations is asserted, not modelled. Critics of associationalism appear only in footnotes.
- Context. Pax Machina's editors and board come largely from frontier labs and AI-policy groups, and Edelman is himself an editor. One footnote cites a suggestion by Claude Fable 5 as a source.
About Pax Machina
Tagline: "Proposals and debate, for institutions in a world with powerful AI." Editors include Séb Krier (Google DeepMind), Ryan Lowe and Oliver Klingefjord; the board includes Iason Gabriel, Saffron Huang, Seth Lazar and Dean W. Ball. Earlier issues include Nick Caputo on AGI's bureaucratic future and Justin Curl on AI courts.